Sign in with
Sign up | Sign in
Your question

3DMark06: Phenom X4 vs C2Q

Last response: in CPUs
Share
November 6, 2007 6:46:54 AM

www.expreview.com/news/hard/2007-11-06/1194333585d6811_...

Not as much of a hammering as the Crysis benchmarks, but still 10% slower than Kentsfield and 15% slower than Yorkfield in the CPU score. Note this is an overclocked Phenom @ 3GHz to create a clock for clock comparison with the C2Qs, release speeds will initially be far slower, 2.4GHz this month, followed by 2.6GHz possibly in December or January.

Sorry Baron, looks like Rahul was wrong. :pfff: 

More about : 3dmark06 phenom c2q

November 6, 2007 9:28:17 AM

Not too bad.
a b à CPUs
November 6, 2007 10:07:18 AM

Still must be an ES - look at the HT link speed.

The Intel isn't much faster and yet has twice or 4 times the cache.

Things are looking up for AMD ... catchup time ... yeah !!
Related resources
November 6, 2007 10:24:39 AM

Single channel and 25MHz slower memory could mean this test was worthless. Scoring that high with those disadvantages is no small task. How did they get 3GHz on X4 phenom at 1.184v?
November 6, 2007 10:41:54 AM

elbert said:
Single channel and 25MHz slower memory could mean this test was worthless. Scoring that high with those disadvantages is no small task. How did they get 3GHz on X4 phenom at 1.184v?


Firstly, it's dual channel. I suggest you read the article, as they have an explanation, as Phenom X4 has dual memory controllers, one for each RAM module. CPU-Z can't report this correctly. As for the memory speed, it's a divider issue @ 3GHz. At 3GHz with DDR2-800 you can only get 3000/7 = 428 or 3000/8 = 375, and the motherboard defaults to the lower RAM speed. Even with the RAM overclocked to 428MHz, how much ground do you expect Phenom to gain? Certainly nowhere near enough to match C2Q.

As for the voltages and overclocking details, you would have to ask around @ xtremesystems for answers, as the OP regularly posts there.
November 6, 2007 11:38:39 AM

The aticle is garbage I read it and still di not understand what they wher talking about.hopefully you read the whole article.Less than 2 weeks to go till we know whats up.I dont think the Intels are more money?,am I wrong,anyway in 2 weeks
November 6, 2007 12:09:24 PM

Did I read that wrong, Phenom is slower clock for clock.
November 6, 2007 12:10:49 PM

epsilon84 said:
Firstly, it's dual channel. I suggest you read the article, as they have an explanation, as Phenom X4 has dual memory controllers, one for each RAM module. CPU-Z can't report this correctly. As for the memory speed, it's a divider issue @ 3GHz. At 3GHz with DDR2-800 you can only get 3000/7 = 428 or 3000/8 = 375, and the motherboard defaults to the lower RAM speed. Even with the RAM overclocked to 428MHz, how much ground do you expect Phenom to gain? Certainly nowhere near enough to match C2Q.

As for the voltages and overclocking details, you would have to ask around @ xtremesystems for answers, as the OP regularly posts there.



It actually looks like a difference of front side bus and cache. The Kentsfield has 4MB, the Yorkfield has 6MB, Phenom has 2MB L2. 790FX has been said to run at up to 500MHz. Also, with HT3, I believe the RAM does run at rated speeds. I think this is the site that had the OverDrive pic that showed the RAM speed at 533MHz(1066DDR). It's just a matter of running the RAM async.

It was also reported by Digitimes I believe that the channel builders have the chips so, hopefully, retail samples will show up at places like this next week. It should be as close or maybe closer with 1066 and 2.6GHz HT.
November 6, 2007 12:12:47 PM

didn't mean to,this was my first time doing a link
November 6, 2007 12:38:30 PM

How about some real world benchmarks.
November 6, 2007 12:50:44 PM

not so bad
November 6, 2007 12:53:40 PM

BaronMatrix said:
Well, according to that link the 9700 is a 2.6GHz chip @ 125W.


From Artical:
"Finally, retailers are also indicating that the AMD Phenom X4 9700 will make its debut with the SKU HD9700XAGDBOX and at a retail price of $330. The Phenom X4 9700 is slightly more high-end compared to the Phenom X4 9600 and Phenom X4 9500, and features a 2.4GHz operating speed along with a 125-Watt Thermal Design Power envelope.

AMD's original roadmap slated the clock frequency for the Phenom 9600 at 2.4 GHz; the original clock frequency for the 9700 was at 2.6 GHz. It might be easy to assume the worst with the sudden decrease in core frequencies, though the change could also be attributed to creative marketing plans."



They changed it.
November 6, 2007 1:13:26 PM

According to the author, it seems like Phenom X4s are poor overclockers. That's why they used ES, because it has unlocked multiplier. For some reason the author continuously hammered this point.
a c 126 à CPUs
November 6, 2007 1:31:13 PM

elbert said:
How did they get 3GHz on X4 phenom at 1.184v?

same way you can with Intel cpu's...it droops bit 1.184 is the average :) 
a b à CPUs
November 6, 2007 1:54:05 PM

For the guys saying it's a good performance.... You *are* aware the Phenom is running a 25% overclock versus stock clocked C2Q's, right? (3/2.4 = 1.25).
November 6, 2007 2:02:37 PM

We can also compare it at stock speed, but what would that result in? We want to know how it compares clock for clock.

They could also underclock the C2Q's.

They are bad OC, well again, only a small part of the people OC there cpu, yes more people here then everywhere else, but still. For the clockers under us, it will be a bad/slightly worst buy.

Ah well, a few more weeks, then we will see real tests, hopefully.
November 6, 2007 2:03:06 PM

also. that is a 500 dollar proc against a 1000 dollar proc (for the qx6850) ten percent less performance for 50 percent less price sounds good.)
November 6, 2007 2:13:40 PM

I hear ya but my $269 Q6600 G0 OC's to 3.6 effortlessly without voltage changes.
November 6, 2007 2:21:01 PM

If this was the other way round and AMD Phenom was winning I'm sure many posters would be complaining 'This is fake'.

They supposedly have a Phenom X4 (no one else does) :??:  , but why only a select few benchmarks ? They decide to select a synthetic 3Dmark Benchmark :sarcastic:  Superpi well that's really useful :sarcastic:  A fair comparison would be a variety of games, encoding tasks etc. I'm going to wait for some trusted sources to do a comparison then decide which is better.
a b à CPUs
November 6, 2007 2:28:20 PM

Phenom should be using DDR2-1066 not DDR2-800.... The Core2's might not care what the speed of the memory is as proven by quite a few benchmarks but looking at the benchmarks on this site for crysis, Phenom seems to care alot about the memory speed. Crysis was given an average of a 8-12.1% AVG 11.7% boost from going from 375Mhz to 508Mhz (DDR2-800 at 3Ghz to DDR2-1066 [sortof] at 3.05Ghz) which if 3DMark06 is as Memory loving as Crysis was for the Phenom the scores would sway in AMD's favor a bit.

But anyways we should wait till we see benchmarks running on AM2+ boards to see what Phenom's performance really is..
November 6, 2007 2:55:48 PM

Scotteq said:
For the guys saying it's a good performance.... You *are* aware the Phenom is running a 25% overclock versus stock clocked C2Q's, right? (3/2.4 = 1.25).

Yes, I think most people are aware some chips are overclocked so they all match at 3000mhz.
I'm just happy to just see the X4 close (within 40%) to the C2Quad. Just for the sake of keeping prices low, making new products, and keeping Intel from being a monopoly.
Not worried about AMD selling chips, The X2 is still very popular in developing markets for good reasons.
November 6, 2007 4:01:08 PM

Since the tested AMD cpu is only an engineering sample and is using DDR2 800 ram instead of DDR2 1066, I think the Phenom came out failry good. Its not great and could use some work, but that's the difference between an engineering sample and a an actual release chip. If memory serves me correctly, the higher end Phenoms will have a 2x1024 L2 cache, not a 2x512. Could be mistaken about that, but if it does have the 2x1024, the results could very well improve.

When all is said and done, I want real benches done with retail chips on a high performance motherboard using decent ram. Then I will make a decission as to how good or bad the Phenom is, and whether its worth my money or not.
November 6, 2007 9:02:47 PM

elbert said:
Single channel and 25MHz slower memory could mean this test was worthless. Scoring that high with those disadvantages is no small task. How did they get 3GHz on X4 phenom at 1.184v?

by ten points maybe not hundreds
November 7, 2007 12:24:29 AM

epsilon84 said:
Firstly, it's dual channel. I suggest you read the article, as they have an explanation, as Phenom X4 has dual memory controllers, one for each RAM module. CPU-Z can't report this correctly. As for the memory speed, it's a divider issue @ 3GHz. At 3GHz with DDR2-800 you can only get 3000/7 = 428 or 3000/8 = 375, and the motherboard defaults to the lower RAM speed. Even with the RAM overclocked to 428MHz, how much ground do you expect Phenom to gain? Certainly nowhere near enough to match C2Q.

As for the voltages and overclocking details, you would have to ask around @ xtremesystems for answers, as the OP regularly posts there.

Thats funny another test with phenom had CPU-z showing dual channel from this site. This link show a test done less than a week ago with the correct dual channel for CPU-z from this site. Strange why they didn't stay with this version of CPU-z. Here I was think they had an unlocked X4 engineering sample where someone could just up the mobo clock to get the 25MHz added.
http://www.expreview.com/news/hard/2007-10-29/1193620830d6602.html
What if AMD suggest DDR2 1066 memory for the X4 and the new AM2+ work with them? IE no OC for 1066. I dont know about 428MHz but I have seen a test with memory at 500MHz and the multiplier lowered to 11x added 8FPS in crysis to the test with memory set to 375MHz.

Lets not forget this is an average of games which isn't well optimized for quad core. IE no where near a 50% advantage going from E6850 to QX6850.
November 7, 2007 1:35:41 AM

shouldn't they try this with ddr800 CAS 3, since aren't Phenoms perforance more related to memory due to less cache?
a b à CPUs
November 7, 2007 2:27:01 AM

Wow Interesting take a close look at those pictures. In Reference to the Epsilon84/Elbert Argument..

http://www.expreview.com/img/news/071028/Crysis_k10.jpg
http://www.expreview.com/img/news/071106/phenom_06.png (Warning Small)

1) Notice that the voltage has dropped by 0.3 volts from 1.536v in the Crysis Pic to 1.184v in the 3Dmark pic.
2) Notice that the Version of CPU-ID is exactly the same.
3) The memory Page went from Dual (Crysis) to Single (3Dmark)

Questions of Wonder
?? Is this the same Engineering Sample??
Are some features locked in the chips? If so is this chip really going to show us an accurate representation of Phenom's actual speed. They both mention using an RD790 chipset... So is this an early bios?

Anyways all we can do is wait 2 weeks and hope that they actually produce themselves along with the RV670 videocards and the RD7x0 Chipsets.
November 7, 2007 5:08:26 AM

I have absolutely no idea how they managed to overclock to 3GHz at that voltage, or even if CPU-Z is reporting the voltage correctly. It could well be a 'power down' idle state for the CPU where voltage is reduced, who knows.

What I do know is that they had some odd peculiarities in their initial overclocking attempt, and they ended up reinstalling the OS and got better results in benchmarks afterwards. This is straight from the mouth of the guy who runs expreview...
November 7, 2007 5:26:13 AM

elbert said:
Thats funny another test with phenom had CPU-z showing dual channel from this site. This link show a test done less than a week ago with the correct dual channel for CPU-z from this site. Strange why they didn't stay with this version of CPU-z. Here I was think they had an unlocked X4 engineering sample where someone could just up the mobo clock to get the 25MHz added.
http://www.expreview.com/news/hard/2007-10-29/1193620830d6602.html
What if AMD suggest DDR2 1066 memory for the X4 and the new AM2+ work with them? IE no OC for 1066. I dont know about 428MHz but I have seen a test with memory at 500MHz and the multiplier lowered to 11x added 8FPS in crysis to the test with memory set to 375MHz.

Lets not forget this is an average of games which isn't well optimized for quad core. IE no where near a 50% advantage going from E6850 to QX6850.


They did a reinstall of the OS during that period. Some people suspect they have more than one ES as well... judging by the variances in voltages and what not. ;) 

I do agree that without DDR2-1066, Phenom is certainly not showing it's full potential. But isn't this more a weakness in the architecture, rather than an advantage, considering the cost difference between DDR2-1066 and DDR2-800? You can get 2GB DDR2-800 for around $50, the same amount for DDR2-1066 costs over $100...

1333FSB C2Ds runs almost just as quick with run of the mill generic CL5 DDR2-667 as it does with CL4 DDR2-1066 - in most benchmarks (except bandwith tests of course) the difference will at most be around 5%, usually even less.

One last thing, in regards to the 3DMark06 scores in general, I don't believe it is very sensitive to memory bandwith/timings, at least compared to most other benchmarks.
November 7, 2007 6:55:11 AM

These benchmarks are bull - nonsense - wait for the real stuff and get back to work - this is getting even worse than an over obsesive compulsive gamer.

Jippee they use a fsb of this and that bla bla blur blur - you num nuts - you're not even qualified to speculate like this - ( for those who know sooo much )

Grow up
November 7, 2007 9:27:07 AM

Wow what a surprise. Barcelona showed K10 is slower c-f-c than Conroe. Not surprising that Phenom is too.

Of course the AMD fan bois will screech none the less.
a b à CPUs
November 7, 2007 10:34:28 AM

MrsB - It's difficult to make a true comparison based on a single measure. Especially on an artificial benchmark rather than more real world performance tasks. But having said that: You *are* aware that the AMD chip is running a 25% overclock in this comparison, right???

Care to revisit your "superior" comment?
November 7, 2007 12:01:33 PM

Scotteq said:
MrsB - It's difficult to make a true comparison based on a single measure. Especially on an artificial benchmark rather than more real world performance tasks. But having said that: You *are* aware that the AMD chip is running a 25% overclock in this comparison, right???

Care to revisit your "superior" comment?



Actually, if it's the same speed as the Barcelona ES, it's at 1.6GHz before the OC. That means that it can take a close to 100% OC. But the point is to test the same clockspeed.
a b à CPUs
November 7, 2007 1:24:55 PM

Baron - It's 1.6? I was under the impression the stock speeds for Phenom would be 2.4, and did the math from there. 2.4 to 3 is a 25% bump.

So: Kudos to AMD for producing a chip that can hit a 100% overclock on air and that kind of voltage. Especially at that die size. That really is impressive. And I'm serious in saying that. Get much past 33% with an Intel chip, and it's time to call the plumber. <grin>

Absolutely, it should do very well at the lower end of the marketplace given the potential cost/performance ratio. It's very good, certainly. But I don't see where we get to "superior" though.
November 7, 2007 2:43:22 PM

epsilon84 said:

I do agree that without DDR2-1066, Phenom is certainly not showing it's full potential. But isn't this more a weakness in the architecture, rather than an advantage, considering the cost difference between DDR2-1066 and DDR2-800? You can get 2GB DDR2-800 for around $50, the same amount for DDR2-1066 costs over $100...

It more out the be forward looking as this design may quickly find itself on 45nm and DDR3. Its kind of like the new games that no system can play at max until a year or 2 later. AMD and Intel has lots of money in design and to only design a CPU to do well on slow low cost memory isn't a good long term investment. AMD and Intels core designs last for about 3 years with a few revisions.
November 7, 2007 2:59:03 PM

Quote:
It looks like a pretty damn good chip from little AMD.
It says a lot that Intel felt they needed 6X the cache to stay ahead of phenoms performance. That tells me AMD's products IS superior, and Intel knew it thats why they had to add sooooo much cache.


...product is superior due to its design choice?
November 7, 2007 3:06:14 PM

BaronMatrix said:
Actually, if it's the same speed as the Barcelona ES, it's at 1.6GHz before the OC. That means that it can take a close to 100% OC. But the point is to test the same clockspeed.


I think the ES is 2.0Ghz rather than 1.6Ghz, but I'm not sure. The author of Expreview also stressed that Phenom is very difficult to overclock, and that's why they used the ES.

Then they reminded us that the overclockability found in ES may not apply to production chip. Sounds like if you can unlock the multiplier, you can OC this thing a bit. But if you can only change the FSB, you're pretty much locked.
November 7, 2007 3:20:15 PM

intel mintel gintel

tranlated

inferior mental gintelia
in short its short
November 7, 2007 3:20:29 PM

Not very bad actually, but no way to say Phenom will be a good performer comparing to 45nm Core 2 Quad...

375MHz memory clock is what Phenom can do at 3GHz with its integrated memory controller, no way to challenge for that 25MHz slower memory clock, because that is what Phenom can do in 3GHz...

Just understand why AMD insists on spending so much time on developing the original 4 cores architecture, it does really double the score clock for clock compare to Brisbane. It just doesn't make sense why border to have K10 architecture develop and what performance advancement does K10 bring to the market. There's no performance gain shown with the addition of K10 advancement, at least for the current result of Phenom, then why border spending so much time for K10?
November 7, 2007 3:50:30 PM

but there is no 3ghz phenom, it's an overclock that made this bad memfreq (true enough that we have seen in the past that they'll produce chips with this bad freq anyway).

But we should see them all on full potential if you ask me and put in the higher speed DDR2/3 rams, make more tests, compare the chip against a few options. Let's see what the effect is of running higher and lower memory.

And they made the K10 more for the server market. MCM has more use for monolith cores. They made a mistake, they saw the mistake Intel made with it's glued cores and thought it wouldn't work again (at least I think that was one of the reasons). The mem problem is a byproduct of the on die mem controller.

I hope the real phenoms perform better, if it really will, we have to wait and see, but at least lets hope it's a good proc or at least good for the price.
November 7, 2007 5:00:37 PM

elbert said:
It more out the be forward looking as this design may quickly find itself on 45nm and DDR3. Its kind of like the new games that no system can play at max until a year or 2 later. AMD and Intel has lots of money in design and to only design a CPU to do well on slow low cost memory isn't a good long term investment. AMD and Intels core designs last for about 3 years with a few revisions.


Fair point on the forward looking part, but I'm talking RIGHT NOW, as in the upcoming launch, due to the neccesity of high speed RAM, the total cost of ownership of Phenom will not be cheap. We're looking at ~$300 for the CPU, ~$100 for the RAM, and initial AM2+ RD790 boards will definitely cost a lot more than current AM2 boards. Prices will settle eventually but we're looking at perhaps $550 - $600 for an AM2+ based Phenom, compared with $450 - $500 for a reasonably specced P35/Q6600 platform.

I do wonder how making a CPU that perform wells *without* needing high end memory not a good long term investment? Are the initial 1066FSB C2Ds that could run perfectly fine with DDR2-533 suddenly 'not a good investment' just because DDR2-667/800 is the 'standard' now? Of course not, they are still within 2 - 3% of 1333FSB C2Ds in most benchmarks.
November 7, 2007 5:48:39 PM

This is bias... since the Phenom X4 was running not on its native clock speed, its like forced labor.

They should compare Phenom X4 (2.4GHz) with Q6600 (2.4Ghz) thats a fair comparison.
November 7, 2007 6:40:30 PM

I just wonder what side of the fence people would be on if Intel was comparing their $1,000 CPU to one from AMD at $300.00??? Instead of the other way around............ QX6850 $1029.99 proposed price for Phenom X4 $300.00

What would give you more bang for your buck???
November 7, 2007 6:48:49 PM

epsilon84 said:
Fair point on the forward looking part, but I'm talking RIGHT NOW, as in the upcoming launch, due to the neccesity of high speed RAM, the total cost of ownership of Phenom will not be cheap. We're looking at ~$300 for the CPU, ~$100 for the RAM, and initial AM2+ RD790 boards will definitely cost a lot more than current AM2 boards. Prices will settle eventually but we're looking at perhaps $550 - $600 for an AM2+ based Phenom, compared with $450 - $500 for a reasonably specced P35/Q6600 platform.

I do wonder how making a CPU that perform wells *without* needing high end memory not a good long term investment? Are the initial 1066FSB C2Ds that could run perfectly fine with DDR2-533 suddenly 'not a good investment' just because DDR2-667/800 is the 'standard' now? Of course not, they are still within 2 - 3% of 1333FSB C2Ds in most benchmarks.

The memory controller on the X4 can work with both ddr2 and ddr3. At what points must the mobo change for new memory or more importantly how does current mobos run that work with both? From what I have seen the mobos supporting both DDR2 and DDR3 perform not as well as only have either DDR2 or DDR3. IE either take a performance hit now but it works for upcomming hardware longer. Most are going with the faster mobo now and have to buy a new mobo for DDR3 support which I guess mobo makers will love. CPU's on the other hand would require major redesign for where the controller is on the die.

There are many other issues such as cache efficiency at a given GHz. IE will the cache and memory controller end up starving the design for data at 5GHz and require 16mb of cache or more. Can the number of cores scale and allow for the L3 cache to remain as 1? Theres is lots of other possibility's that new core designs must accomplish to give them the edge over their competition.
November 7, 2007 6:54:16 PM

elbert said:
The memory controller on the X4 can work with both ddr2 and ddr3. At what points must the mobo change for new memory or more importantly how does current mobos run that work with both? From what I have seen the mobos supporting both DDR2 and DDR3 perform not as well as only have either DDR2 or DDR3. IE either take a performance hit now but it works for upcomming hardware longer. Most are going with the faster mobo now and have to buy a new mobo for DDR3 support which I guess mobo makers will love. CPU's on the other hand would require major redesign for where the controller is on the die.

There are many other issues such as cache efficiency at a given GHz. IE will the cache and memory controller end up starving the design for data at 5GHz and require 16mb of cache or more. Can the number of cores scale and allow for the L3 cache to remain as 1? Theres is lots of other possibility's that new core designs must accomplish to give them the edge over their competition.


Can you list the AM2 motherboards that will run with DDR3 memory?
I looked at Newegg and found 0.
a b à CPUs
November 7, 2007 7:21:42 PM

Little said you have to overclock the amd quad to barely beat the Intel E6850 dual. But still gets stomped by Intel quads (that's are NOT overclocked).
November 8, 2007 1:00:25 AM

NMDante said:
Can you list the AM2 motherboards that will run with DDR3 memory?
I looked at Newegg and found 0.

I was talking about Intel mobo's having both DDR2 and DDR3.
November 8, 2007 2:35:33 AM

elbert said:
I was talking about Intel mobo's having both DDR2 and DDR3.


Ah, okay.

You started talking about X4 memory controller, so I assumed you were talking about AMD motherboards.

!